We have developed an algorithm for inverting broadband electromagnetic (EM) data to determine its electrical conductivity, magnetic permeability, radius, and depth using a sphere model. The algorithm is based on traditional non-linear inverse methods with singular value decomposition. Studies using synthetic GEM-3 broadband EM data indicate that the technique is dependable and produces fastconverging solutions. The geometrical parameters obtained through this technique are very close to true values, and the electric parameters for noise-free data are correct to within a few percent. When a 10 percent noise is added to the data, the geometric parameters are still well determined, but the electric parameters are poorly resolved, particularly for a sphere that is both conductive and permeable. We also present an analysis of the relative importance and reliability of the model parameters based on their partial derivatives with respect to EM response.